Read and Load the dataset¶

In [1]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

# Load the specific sheet named 'Webinar Leads' from the Excel file
df = pd.read_excel('US Pathway Complete Data.xlsx', sheet_name='campaign_performance')

# Display the first few rows of the dataframe
df.head()
Out[1]:
dates campaign_name campaign_start_date creative_name total_spent impressions clicks click_through_rate leads platform adset_name
0 2024-04-29 USP_Search_APTNTSKA_250124 NaT -- 6766.15 1250 220 0.176000 25.0 Google GMAT/GRE
1 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 NaT #7 ( Graphic ) - Get A Free Pass 946.38 3154 9 0.285352 1.0 Facebook USP-2_KA_TN_050424
2 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 NaT USP_S3_never-been-easier 170.40 2695 8 0.296846 0.0 Facebook USP-2_KA_TN_050424
3 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 NaT USP_S4_Do-you-have 55.65 577 2 0.346620 0.0 Facebook USP-2_KA_TN_050424
4 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 NaT USP_S1_We-guarantee 296.94 1466 10 0.682128 1.0 Facebook USP-2_AP_TS_050424
In [2]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 28534 entries, 0 to 28533
Data columns (total 11 columns):
 #   Column               Non-Null Count  Dtype         
---  ------               --------------  -----         
 0   dates                28534 non-null  datetime64[ns]
 1   campaign_name        28534 non-null  object        
 2   campaign_start_date  4247 non-null   datetime64[ns]
 3   creative_name        28534 non-null  object        
 4   total_spent          28534 non-null  float64       
 5   impressions          28534 non-null  int64         
 6   clicks               28534 non-null  int64         
 7   click_through_rate   28497 non-null  float64       
 8   leads                28534 non-null  float64       
 9   platform             28534 non-null  object        
 10  adset_name           24287 non-null  object        
dtypes: datetime64[ns](2), float64(3), int64(2), object(4)
memory usage: 2.4+ MB

Handling null's¶

In [3]:
# Check the data type of 'leads' before conversion
print(df['leads'].dtype)

# Check for non-numeric values and NaNs
print(df['leads'].unique())

# Fill NaNs with 0 or another value if necessary
df['leads'] = df['leads'].fillna(0)

# Convert 'leads' to integer
df['leads'] = df['leads'].astype(np.int64)

# Check the data type of 'leads' after conversion
print(df['leads'].dtype)

float64
[ 25.     1.     0.     6.     4.     2.     8.     3.    11.     7.
  14.     5.    10.    15.    20.    28.     9.    16.    23.    22.
  18.    38.    31.    16.91  40.    13.    12.    19.     6.28  36.
   4.09  51.    17.    33.    37.    24.    30.    46.    45.    26.
  21.68   8.76  21.   128.    29.    13.24  43.    15.5   11.48   2.69
  14.98   7.52  27.    32.     4.5    0.18   9.36  72.    39.    34.
   0.98  52.    50.    59.    66.     2.03   7.5   55.     3.5    1.78
  35.    73.     9.5    1.19   6.81  10.86  12.14 103.    53.     0.5
   1.82  23.18   2.64   3.64   3.67  19.83  44.     8.13  12.68   3.52
  41.     5.33   0.97   2.61  17.66 153.    16.36  81.    80.     5.5
  29.22   8.16  64.    25.67  17.56  13.68  54.   114.     4.29   1.37
   3.88  11.63  98.     5.78  14.5   42.    14.89   3.87   0.64   2.52
  21.5    1.5    1.64   1.36]
int64

Leads will be mostly the count of people ,so the datatype must be in the form of integer.

In [4]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 28534 entries, 0 to 28533
Data columns (total 11 columns):
 #   Column               Non-Null Count  Dtype         
---  ------               --------------  -----         
 0   dates                28534 non-null  datetime64[ns]
 1   campaign_name        28534 non-null  object        
 2   campaign_start_date  4247 non-null   datetime64[ns]
 3   creative_name        28534 non-null  object        
 4   total_spent          28534 non-null  float64       
 5   impressions          28534 non-null  int64         
 6   clicks               28534 non-null  int64         
 7   click_through_rate   28497 non-null  float64       
 8   leads                28534 non-null  int64         
 9   platform             28534 non-null  object        
 10  adset_name           24287 non-null  object        
dtypes: datetime64[ns](2), float64(2), int64(3), object(4)
memory usage: 2.4+ MB
In [5]:
df['creative_name'].unique()
Out[5]:
array([' --', '#7 ( Graphic ) - Get A Free Pass',
       'USP_S3_never-been-easier', 'USP_S4_Do-you-have',
       'USP_S1_We-guarantee', 'USP_kavita&steffi_VideoAd6',
       '#8 ( Graphic ) - Study In The USA', 'USP_S2_Faced-rejection',
       'USP_S5_easy-for-you', 'State University Of New York_New',
       'USP_RG_VideoAd5', '#15 ( Graphic ) - Dual coursework_Plain',
       "#18 - Master's Degree In 12 Months – Rutgers",
       '#16 ( Graphic ) - Dual coursework_Hands',
       "#6 ( Graphic ) - Master's Degree In 12 Months",
       'Tomorrow_Static-6', 'USP_shreya&steffi_VideoAd8',
       'USP_Drexel_1_EWYL', 'USP_shreya_VideoAd7',
       'USP_Drexel_EWYL_new-Webinar', '15-days left_Static-1',
       '#4 ( Graphic ) - How To Get Admmission In Your Dream Universities In US',
       "#5 ( Graphic ) - Get Master's Degree in Just 12 Months",
       'less than a week_Static-3', 'USP_Drexel_Repay_new-Webinar',
       'USP_Drexel_EM_new-Webinar', 'USP_S12_Rutgers2',
       'USP_S14_Akron_13L', 'USP_S13_Akron', 'USP_S11_Rutgers',
       'USP_S10_Rider_13L', 'USP_S9_Stony_25L', '10-days left_Static-2',
       'USP_S6_SeattleU', '5 days left_Static-5', 'Last chance_Static-4',
       'USP_S8_Rider', 'USP_RIT_1_Getpaid', 'USP_RIT_2_EWYL_INR',
       'USP_S7_Stony', 'USP_Drexel_EWYL_new', 'USP_Drexel_EM_new',
       'USP_Drexel_Repay_new', 'USP_Webinar_without_GMAT/GRE',
       'USP_Program_Linkedin_Static_C2_Cost-Generic',
       'USP_We-guarantee_C32', 'USP_never-been-easier_C34',
       'USP_Program_Linkedin_Static_C12_CW-3YWP', 'USP_easy-for-you_C36',
       'USP_RG_webinar_V1', 'USP_Do-you-have_C35',
       'USP_Program_Linkedin_Static_C11_SB-3YWP',
       'USP_Program_Linkedin_Static_C28_CWRU_3YWP',
       'USP_Program_Linkedin_Static_C3_CWRU-Cost',
       'USP_Program_Linkedin_Static_C13_Notepad',
       'USP_Program_Linkedin_Static_C14_Bridge-3YWP',
       'USP_Program_Linkedin_Static_C19_Dual-Course-white',
       'USP_Shreya_webinar_V1', 'USP_Influencer_videoAd4',
       'Steffi and Shreya', 'USP_Program_Linkedin_Static_C5_Rutgers-Cost',
       'USP_Program_Linkedin_Static_C31_2.25L',
       '#2 ( Graphic ) - State University Of New York',
       '#3 ( Graphic ) - Worried About GMAT Score?',
       '#1 ( Graphic ) - Want To Build Your Career In The US?',
       'USP_Faced-rejection_C33',
       'USP_Program_Linkedin_Static_C9_Avg-Salary',
       'USP_Webinar_Saturdayfuture',
       "#11 ( Graphic ) - Master's Degree In 12 Months_CWRU",
       'USP_RG_webinar_V4_Namita', '#9 ( Graphic ) - Study In The US',
       '#10 ( Graphic ) - Get Into Top US Universitie',
       '#17 ( Graphic ) 2.25L',
       'USP_Program_Linkedin_Static_C7_Drexel-Cost',
       'USP_Influencer_videoAd2', 'USP_Influencer_videoAd3',
       'USP_Influencer_videoAd1', '#9 ( Graphic ) - Study In The USA',
       'USP_Program_Linkedin_Static_C17_SUNY-Cost-white',
       'USP_Program_Linkedin_Static_C15_RIT-Cost-white',
       'USP_Program_Linkedin_Static_C27_Dual-coursework_Hands',
       '#7 ( Graphic ) - Get a free pass',
       '#8 ( Graphic ) - Study in The USA',
       'USP_Program_Linkedin_Static_C24_SB-22L',
       'USP_Program_Linkedin_Static_C4_RIT-Cost',
       '#9 ( Graphic ) - Study in The USA',
       'USP_Program_Linkedin_Static_C29_Rutgres_3YWP', 'Shreya',
       'USP_Program_Linkedin_Static_C23_Rutgers-33K', 'RG Video',
       'USP_Program_Linkedin_Static_C1_Without-Any-Barriers',
       'USP_Program_Linkedin_Static_C25_Tulane-19L',
       'USP_Program_Linkedin_Static_C8_Dual-Course',
       'USP_Program_Linkedin_Static_C30_Tulane_3YWP',
       'USP_Fairfield-MBA_2', 'USP_RG_webinar_V2', 'Kavita and Steffi',
       'USP_Program_Linkedin_Static_C22_Drexel-27L',
       'USP_Program_Linkedin_Static_C26_Dual-coursework_plain',
       'USP_RG_webinar_V3_Mayank',
       'USP_Program_Linkedin_Static_C16_Rutgers-Cost-white',
       'USP_Drexel_6_MSCE_USD',
       'USP_Program_Linkedin_Static_C10_SB-Savings',
       'USP_Webinar_without_GMAT/GRE - Copy', 'USP_Generic-MBA_3',
       '#7 ( Graphic ) - Get free pass',
       'USP_Program_Linkedin_Static_C21_CWRU-22L',
       'USP_Webinar_SaturdayPlan_1', 'USP_Webinar_MSunder12Months - Copy',
       'USP_Webinar_MSunder12Months', 'Influencer 3', 'Influencer 4',
       "#14 ( Graphic ) - Master's Degree In 12 Months_Drexel",
       'USP_Webinar_SaturdayPlan_1 – Copy',
       'USP_Webinar_SaturdayPlan_1 – Copy 2', 'USP_RIT_2_Repay',
       "#13 ( Graphic ) - Master's Degree In 12 Months_Tulane",
       "#12 ( Graphic ) - Master's Degree In 12 Months_Rutgers",
       'USP_Drexel_5_MSIOT_INR', 'USP_Drexel_2_EM', 'USP_Drexel_4_Repay',
       'USP_Drexel_8_MSIOT_USD', 'USP_Drexel_7_MSCE_INR',
       'USP_Drexel_3_Getpaid', 'USP_Program_Linkedin_Static_C6_SUNY-Cost',
       'USP_Webinar_Saturdayfuture - Copy', 'Influencer 1',
       'USP_RIT_1_Repay', 'USP_Drexel-MBA_1',
       'USP_Fairfield_MS-Marketing-2_5', 'USP_RIT_2_Getpaid',
       'USP_RIT_3_EWYL_INR', 'USP_RIT_4_EWYL_USD',
       'USP_Fairfield_MS-Marketing_4',
       '#9 ( Graphic ) - Study In The US – Copy',
       'USP_Program_Linkedin_Static_C20_CWRU-Cost-white',
       'USP_Program_Linkedin_Static_C20_CWRU-Cost-White',
       'USP_Program_Linkedin_Static_C18_Drexel-Cost-white',
       "#6 ( Graphic ) - Master's Degree In 12 Months – Copy",
       '#1 ( Graphic ) - Want To Build Your Career In The US? – Copy',
       '#3 ( Graphic ) - Worried About GMAT Score? – Copy',
       '#8 ( Graphic ) - Study In The USA – Copy',
       '#2 ( Graphic ) - State University Of New York – Copy',
       'Influencer 2', "#19 - Master's Degree In 12 Months – CW",
       'USP_Drexel_1_Getpaid'], dtype=object)

Based on our observation in Power BI, we noticed that the campaign names for two specific campaigns (USP_Search_APTNTSKA_250124 and USP_Search_India-ex-APTNTSKA_250124) were displayed as '--'. Given the importance of maintaining data integrity and the inability to accurately assign them as campaign names or adset names, we recommend leaving these entries as they are. This approach will ensure the accuracy and consistency of the dataset without introducing any assumptions or errors.

In [6]:
df.isnull().sum()
Out[6]:
dates                      0
campaign_name              0
campaign_start_date    24287
creative_name              0
total_spent                0
impressions                0
clicks                     0
click_through_rate        37
leads                      0
platform                   0
adset_name              4247
dtype: int64
In [7]:
# Drop the 'campaign_start_date' column
df = df.drop(columns=['campaign_start_date'])

We have removed the campaign_start_date column from the dataset because it contained a significant number of null values, making it impractical to fill in accurately. Since the column was not useful for our Exploratory Data Analysis (EDA), omitting it will streamline our analysis and focus on more relevant data.

In [8]:
df.isnull().sum()
Out[8]:
dates                    0
campaign_name            0
creative_name            0
total_spent              0
impressions              0
clicks                   0
click_through_rate      37
leads                    0
platform                 0
adset_name            4247
dtype: int64
In [9]:
# Fill null values with 0 in the 'click_through_rate' column
df['click_through_rate'] = df['click_through_rate'].fillna(0)

We filled the null values in the click_through_rate column with '0'. This decision was based on our observation in Power BI that these rows had no clicks. Therefore, it is logical and accurate to set the click_through_rate to '0' for those entries. This ensures the integrity of the dataset and allows for more accurate analysis.

In [10]:
df.isnull().sum()
Out[10]:
dates                    0
campaign_name            0
creative_name            0
total_spent              0
impressions              0
clicks                   0
click_through_rate       0
leads                    0
platform                 0
adset_name            4247
dtype: int64
In [11]:
# Fill null values in 'adset_name' with '--'
df['adset_name'] = df['adset_name'].fillna('--')

The null values in the adset_names column are due to the absence of ad set names for all campaigns on LinkedIn. Therefore, we cannot fill these values with creative_name or campaign_name. Instead, we will retain them as '--' to maintain the accuracy and consistency of the dataset.

In [12]:
df.isnull().sum()
Out[12]:
dates                 0
campaign_name         0
creative_name         0
total_spent           0
impressions           0
clicks                0
click_through_rate    0
leads                 0
platform              0
adset_name            0
dtype: int64

All null values were cleared.

In [13]:
df.head()
Out[13]:
dates campaign_name creative_name total_spent impressions clicks click_through_rate leads platform adset_name
0 2024-04-29 USP_Search_APTNTSKA_250124 -- 6766.15 1250 220 0.176000 25 Google GMAT/GRE
1 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 #7 ( Graphic ) - Get A Free Pass 946.38 3154 9 0.285352 1 Facebook USP-2_KA_TN_050424
2 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S3_never-been-easier 170.40 2695 8 0.296846 0 Facebook USP-2_KA_TN_050424
3 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S4_Do-you-have 55.65 577 2 0.346620 0 Facebook USP-2_KA_TN_050424
4 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S1_We-guarantee 296.94 1466 10 0.682128 1 Facebook USP-2_AP_TS_050424

Handling Outliers¶

In [14]:
# Set the style for the plots
sns.set(style="whitegrid")

# List of numerical columns to plot
columns_to_plot = ['total_spent', 'impressions', 'clicks', 'leads']

# Create a figure and axes for subplots
fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(18, 12))
axes = axes.flatten()

# Plot box plots for each numerical column
for i, column in enumerate(columns_to_plot):
    sns.boxplot(data=df, x=column, ax=axes[i])
    axes[i].set_title(f'Box Plot of {column}')
    axes[i].set_xlabel(column)

plt.tight_layout()
plt.show()

Based on our analysis of the data and graphs, we observed a skewed distribution, indicating the presence of outliers. Although the Interquartile Range (IQR) method is a standard approach for handling outliers, applying it in this case would result in the removal of a significant portion of the data. Considering the client's requirements and the necessity to retain all data points for comprehensive analysis, we recommend not removing any outliers in this dataset. This approach will ensure that the integrity of the dataset is maintained while fulfilling the client's needs.

EDA¶

Univariate Analysis¶

In [15]:
# Dates
plt.figure(figsize=(10, 6))
sns.histplot(df['dates'], bins=30, kde=True)
plt.title('Distribution of Dates')
plt.xlabel('Dates')
plt.ylabel('Frequency')
plt.xticks(rotation=45)
plt.show()

Distribution of dates across the dataset.we can see that these were mostly high in the months of December,February and April.

Distribution of Numerical columns

In [16]:
# Set the plot style
sns.set(style="whitegrid")

# Plot the distribution of total_spent
plt.figure(figsize=(10, 6))
sns.histplot(df['total_spent'], bins=30, kde=True)
plt.title('Distribution of Total Spent')
plt.xlabel('Total Spent')
plt.ylabel('Frequency')
plt.show()

print("Mean : ",df['total_spent'].mean())
print("Median : ",df['total_spent'].median())
print("Min : ",df['total_spent'].min())
print("Max : ",df['total_spent'].max())

Mean :  554.6046712693629
Median :  53.285
Min :  0.0
Max :  27125.36

This skewness towards lower total spent suggests that most of the campaigns or data points in our dataset have relatively fewer total spent, with a concentration of values between 100 and 1000.

In [17]:
# Plot the distribution of impressions
plt.figure(figsize=(10, 6))
sns.histplot(df['impressions'], bins=30, kde=True)
plt.title('Distribution of Impressions')
plt.xlabel('Impressions')
plt.ylabel('Frequency')
plt.show()


print("Mean : ",df['impressions'].mean())
print("Median : ",df['impressions'].median())
print("Min : ",df['impressions'].min())
print("Max : ",df['impressions'].max())

Mean :  2381.018399102825
Median :  107.0
Min :  0
Max :  196585

This skewness towards lower impressions suggests that most of the campaigns or data points in our dataset have relatively fewer impressions, with a concentration of values between 100 and 1000.

In [18]:
# Plot the distribution of clicks
plt.figure(figsize=(10, 6))
sns.histplot(df['clicks'], bins=30, kde=True)
plt.title('Distribution of Clicks')
plt.xlabel('Clicks')
plt.ylabel('Frequency')
plt.show()

print("Mean : ",df['clicks'].mean())
print("Median : ",df['clicks'].median())
print("Min : ",df['clicks'].min())
print("Max : ",df['clicks'].max())

Mean :  18.770624518118737
Median :  1.0
Min :  0
Max :  3361

This skewness towards lower clicks suggests that most of the campaigns or data points in our dataset have relatively fewer clicks, with a concentration of values between 0 and 100.

In [19]:
# Plot the distribution of leads
plt.figure(figsize=(10, 6))
sns.histplot(df['leads'], bins=30, kde=True)
plt.title('Distribution of Leads')
plt.xlabel('Leads')
plt.ylabel('Frequency')
plt.show()

print("Mean : ",df['leads'].mean())
print("Median : ",df['leads'].median())
print("Min : ",df['leads'].min())
print("Max : ",df['leads'].max())

Mean :  1.1915258989275952
Median :  0.0
Min :  0
Max :  153

This skewness towards lower leads suggests that most of the campaigns or data points in our dataset have relatively fewer impressions, with a concentration of values between 0 and 5.We should improve the campaigns with '0' leads or spend less on them.

Distribution of categorical columns

In [20]:
# Set the style for the plot
sns.set(style="whitegrid")

# Create a figure for the plot
plt.figure(figsize=(10, 6))

# Plot the distribution of the 'platform' column
sns.countplot(y=df['platform'], order=df['platform'].value_counts().index)

# Set the title and labels
plt.title('Distribution of Platform')
plt.xlabel('Count')
plt.ylabel('Platform')

# Show the plot
plt.show()

The dataset shows that most of the data points are from Facebook.This suggests that Facebook dominates the dataset, potentially influencing the overall trends and insights derived from the data.

In [21]:
# Set the plot style
sns.set(style="whitegrid")

# Plot the distribution of campaign_name
plt.figure(figsize=(14, 10))
campaign_name_order = df['campaign_name'].value_counts().index
ax = sns.countplot(y=df['campaign_name'], order=campaign_name_order)
plt.title('Distribution of Campaign Names', fontsize=16)
plt.xlabel('Frequency', fontsize=14)
plt.ylabel('Campaign Name', fontsize=14)

# Rotate the labels for better visibility
ax.set_yticklabels(ax.get_yticklabels(), rotation=0, fontsize=13)

plt.show()

The distribution of campaign names indicates that campaigns with more data points generally have higher reach. This is inferred from the fact that campaigns with a larger number of impressions tend to have a higher frequency of occurrences in the dataset. Therefore, campaigns with more data points are likely to have achieved broader reach or engagement.

Bivariate Analysis¶

In [22]:
# Select the columns to analyze
columns_to_analyze = ['impressions', 'clicks', 'leads','total_spent']

# Compute the correlation matrix
correlation_matrix = df[columns_to_analyze].corr()

# Plot the heatmap
plt.figure(figsize=(8, 6))
sns.heatmap(correlation_matrix, annot=True, cmap='coolwarm', fmt='.2f', linewidths=0.5, vmin=-1, vmax=1)
plt.title('Correlation Heatmap between Impressions, Clicks, and Leads')
plt.show()

The correlation matrix provides a rough overview of the relationships between the columns in our dataset. Eg.Impressions and cliks have strong +ve correlation,clicks and leads have weak +ve correlation,impressions and leads have moderately +ve correlation,impressions and total_spent have moderately +ve correlation,clicks and total_spent have moderate +ve correlation,leads and total_spent have moderate +ve correlation.

In [23]:
# Define the pairs of columns to create scatter plots for
pairs = [('impressions', 'clicks'), 
         ('impressions', 'leads'), 
         ('impressions', 'total_spent'), 
         ('clicks', 'leads'), 
         ('clicks', 'total_spent'), 
         ('leads', 'total_spent')]

# Create a 3x2 grid of scatter plots
fig, axes = plt.subplots(3, 2, figsize=(15, 18))

for ax, (x, y) in zip(axes.flatten(), pairs):
    sns.scatterplot(x=df[x], y=df[y], ax=ax)
    ax.set_title(f'{x.capitalize()} vs. {y.capitalize()}')
    ax.set_xlabel(x.capitalize())
    ax.set_ylabel(y.capitalize())

plt.tight_layout()
plt.show()

By visualizing these relationships, the scatterplots provide a graphical confirmation of the correlations observed in the matrix, offering a more intuitive understanding of the data dynamics.

In [24]:
df.head()
Out[24]:
dates campaign_name creative_name total_spent impressions clicks click_through_rate leads platform adset_name
0 2024-04-29 USP_Search_APTNTSKA_250124 -- 6766.15 1250 220 0.176000 25 Google GMAT/GRE
1 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 #7 ( Graphic ) - Get A Free Pass 946.38 3154 9 0.285352 1 Facebook USP-2_KA_TN_050424
2 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S3_never-been-easier 170.40 2695 8 0.296846 0 Facebook USP-2_KA_TN_050424
3 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S4_Do-you-have 55.65 577 2 0.346620 0 Facebook USP-2_KA_TN_050424
4 2024-05-02 USP_2_Lead-gen-2_Top4-states_050424 USP_S1_We-guarantee 296.94 1466 10 0.682128 1 Facebook USP-2_AP_TS_050424
In [25]:
# Convert 'dates' to datetime format if not already done
df['dates'] = pd.to_datetime(df['dates'])

# Setting plot style
sns.set(style="whitegrid")

# Plotting each metric over time in separate subplots in a 2x2 grid
fig, axs = plt.subplots(2, 2, figsize=(14, 12))

# Total Spent
sns.lineplot(ax=axs[0, 0], x='dates', y='total_spent', data=df, marker='o', color='b')
axs[0, 0].set_title('Total Spent Over Time')
axs[0, 0].set_xlabel('Date')  # Added x-axis label
axs[0, 0].set_ylabel('Total Spent')
axs[0, 0].grid(True)

# Impressions
sns.lineplot(ax=axs[0, 1], x='dates', y='impressions', data=df, marker='o', color='orange')
axs[0, 1].set_title('Impressions Over Time')
axs[0, 1].set_xlabel('Date')  # Added x-axis label
axs[0, 1].set_ylabel('Impressions')
axs[0, 1].grid(True)

# Clicks
sns.lineplot(ax=axs[1, 0], x='dates', y='clicks', data=df, marker='o', color='green')
axs[1, 0].set_title('Clicks Over Time')
axs[1, 0].set_xlabel('Date')  # Added x-axis label
axs[1, 0].set_ylabel('Clicks')
axs[1, 0].grid(True)

# Leads
sns.lineplot(ax=axs[1, 1], x='dates', y='leads', data=df, marker='o', color='red')
axs[1, 1].set_title('Leads Over Time')
axs[1, 1].set_xlabel('Date')  # Added x-axis label
axs[1, 1].set_ylabel('Leads')
axs[1, 1].grid(True)

# Adjusting layout to avoid overlap
plt.tight_layout()
plt.show()

Distribution of all the key metrics numerical columns according to the dates or time period using line plot ,and gives the observation of relations between them. We can see they mostly aligned.

In [26]:
plt.figure(figsize=(14, 6))
sns.lineplot(x='dates', y='click_through_rate', data=df, marker='o', color='purple')
plt.title('Click-Through Rate (CTR) Over Time')
plt.xlabel('Date')
plt.ylabel('CTR')
plt.grid(True)
plt.show()
In [27]:
# Convert 'dates' to datetime format if not already done
df['dates'] = pd.to_datetime(df['dates'])

# Determine the top 4 creative_names by average click_through_rate
top_creative_names = df.groupby('creative_name')['click_through_rate'].mean().nlargest(4).index

# Filter the dataframe to only include the top 4 creative_names
df_top4 = df[df['creative_name'].isin(top_creative_names)]

# Setting plot style
sns.set(style="whitegrid")

# Plotting click_through_rate over time, faceted by creative_name
g = sns.FacetGrid(df_top4, col="creative_name", col_wrap=2, height=4, aspect=1.5)
g.map(sns.lineplot, "dates", "click_through_rate", marker="o")

# Adding titles and labels
g.set_titles("{col_name}")
g.set_axis_labels("Date", "Click Through Rate")
for ax in g.axes.flat:
    for label in ax.get_xticklabels():
        label.set_rotation(45)
        
# Adjusting layout to avoid overlap
plt.tight_layout()
plt.show()

The line plots show the click-through rates over time for the top 4 creative names. Each subplot reveals the performance trends for these creatives, highlighting periods of higher or lower click-through rates and allowing for comparison between them. This visualization helps identify which creative names perform consistently well or show notable fluctuations.

In [28]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 28534 entries, 0 to 28533
Data columns (total 10 columns):
 #   Column              Non-Null Count  Dtype         
---  ------              --------------  -----         
 0   dates               28534 non-null  datetime64[ns]
 1   campaign_name       28534 non-null  object        
 2   creative_name       28534 non-null  object        
 3   total_spent         28534 non-null  float64       
 4   impressions         28534 non-null  int64         
 5   clicks              28534 non-null  int64         
 6   click_through_rate  28534 non-null  float64       
 7   leads               28534 non-null  int64         
 8   platform            28534 non-null  object        
 9   adset_name          28534 non-null  object        
dtypes: datetime64[ns](1), float64(2), int64(3), object(4)
memory usage: 2.2+ MB
In [29]:
# Plot total_spent by adset_name
plt.figure(figsize=(14, 6))
sns.barplot(x='campaign_name', y='total_spent', data=df, palette='viridis')
plt.title('Total Spent by campaign')
plt.xlabel('Campaign Name')
plt.ylabel('Total Spent')
plt.xticks(rotation=90)
plt.show()

This analysis displays the campaigns with the highest expenditure. By examining these plots, we can identify which campaigns have the greatest financial investment, indicating where the most money is spent.

In [30]:
# Plot leads by campaign_name
plt.figure(figsize=(14, 6))
sns.barplot(x='campaign_name', y='leads', data=df, palette='magma')
plt.title('Leads by Campaign Name')
plt.xlabel('Campaign Name')
plt.ylabel('Leads')
plt.xticks(rotation=90)
plt.show()

This analysis shows the campaigns and their associated leads. By visualizing this data, we can identify which campaigns are generating the most leads and assess their effectiveness in driving lead generation.

In [31]:
# Aggregate data by campaign name and calculate total spent, impressions, clicks, and leads
campaign_metrics1 = df.groupby(['campaign_name','platform'])[['total_spent', 'impressions', 'clicks', 'leads']].sum().reset_index()

# Sort the data by leads in descending order and get the top 10 campaigns
top_10_campaigns = campaign_metrics1.sort_values(by='leads', ascending=False).head(10)



# Plot the top 10 campaigns by leads
plt.figure(figsize=(14, 8))
sns.barplot(x='leads', y='campaign_name', data=top_10_campaigns, palette='viridis')
plt.title('Top 10 Campaigns by Leads')
plt.xlabel('Leads')
plt.ylabel('Campaign Name')
plt.show()

The analysis highlights the top 10 campaigns that generated the most leads and their distribution. By examining this data, we can identify which campaigns were most successful in lead generation and observe how leads are distributed across these top-performing campaigns.As their total spent is mostly alligned with the leads.

In [32]:
# Display details of the top 10 campaigns
print("Details of Top 10 Campaigns by Leads:")
print(top_10_campaigns)

Details of Top 10 Campaigns by Leads:
                                        campaign_name  platform  total_spent  \
9                     USP Study Abroad_Generic_240224  Facebook   2126684.24   
13                      USP_MS + Coop Campaign_240224  Facebook   1501040.54   
14        USP_Marketing+MBA_New Leads Campaign_240224  Facebook   1125425.52   
10                            USP_2_Lead-gen-1_220124  Facebook    484768.66   
32      USP_Video_More_Volume_No Logic_India_New Link  Facebook    198416.23   
34                                 USP_Webinar_070324  Facebook    752657.01   
24                         USP_Search_APTNTSKA_250124    Google    789571.12   
0   CBO - USP Masterclass Cold [ 15th Dec ] - High...  Facebook    186700.48   
35                     US_Pathway_2_Lead_gen_1_091123  LinkedIn   1884063.88   
12  USP_COOP-Drexel_More_Volume_No Logic_India_240124  Facebook    126657.36   

    impressions  clicks  leads  
9       9265548   48531   4630  
13      4506382   27762   4446  
14      8106285   37727   2390  
10      1668181   10480   2124  
32      1246514   12420   1866  
34      4125312   16957   1616  
24       178830   21663   1358  
0        610938   12853   1346  
35      5751255   34923   1232  
12       395467    5238   1096

This information provides a comprehensive view of the performance of the top campaigns, showing how investment (total spend) correlates with engagement (clicks) and effectiveness (leads).

In [33]:
# Group by adset_name and sum the total_spent
total_spent_by_adset = df.groupby('adset_name')['total_spent'].sum().reset_index()

# Plotting the total_spent by adset_name
plt.figure(figsize=(14, 6))
sns.barplot(x='adset_name', y='total_spent', data=total_spent_by_adset, palette='viridis')
plt.title('Total Spent by Adset Name')
plt.xlabel('Adset Name')
plt.ylabel('Total Spent')
plt.xticks(rotation=90)
plt.show()

The bar chart displays total spend across different ad sets, highlighting which ad sets received the most investment and allowing for easy comparison of spending levels.

In [34]:
# Plot impressions by adset_name
plt.figure(figsize=(14, 6))
sns.barplot(x='adset_name', y='impressions', data=df, palette='viridis')
plt.title('Impressions by Adset')
plt.xlabel('Adset Name')
plt.ylabel('Impressions')
plt.xticks(rotation=90)
plt.show()

The chart shows the distribution of impressions across different ad sets, revealing which ad sets generated the most impressions and allowing for comparison of their reach.The chart reveals variations in reach, helping identify which ad sets were more effective in capturing audience attention.

In [35]:
# Plot leads by adset_name
plt.figure(figsize=(14, 6))
sns.barplot(x='adset_name', y='leads', data=df, palette='viridis')
plt.title('Leads by Adset')
plt.xlabel('Adset Name')
plt.ylabel('Leads')
plt.xticks(rotation=90)
plt.show()

The chart shows the distribution of leads across different ad sets, highlighting which ad sets generated the most leads. It allows for comparison of lead generation performance between ad sets, revealing which ones were most effective at converting impressions into leads. This insight can help evaluate the effectiveness of various ad sets and inform strategies to optimize lead generation efforts.

In [36]:
# Aggregate data by adset_name and calculate total spent, impressions, clicks, and leads
adset_metrics = df.groupby(['adset_name', 'campaign_name','platform'])[['total_spent', 'impressions', 'clicks', 'leads']].sum().reset_index()

# Sort the data by leads in descending order and get the top 20 ad sets
top_20_adsets = adset_metrics.sort_values(by='leads', ascending=False).head(20)


# Plot the top 20 ad sets by leads
plt.figure(figsize=(14, 8))
sns.barplot(x='leads', y='adset_name', data=top_20_adsets, palette='viridis')
plt.title('Top 20 Ad Sets by Leads')
plt.xlabel('Leads')
plt.ylabel('Ad Set Name')
plt.xticks(rotation=90)
plt.show()

The chart displays the top 20 ad sets ranked by lead generation, showcasing which ad sets were the most effective at producing leads. It enables a clear comparison of lead performance among these top ad sets and helps identify which ones delivered the best results. This insight can guide future marketing strategies and optimize lead generation efforts.

In [37]:
# Display details of the top 20 ad sets
print("Details of Top 20 Ad Sets by Leads:")
print(top_20_adsets)

Details of Top 20 Ad Sets by Leads:
                                            adset_name  \
73                 MS + Coop All India Ad set - 240224   
5             All India Counselling Look-a-like_290224   
116        USP-2_Data_Job-title_India_exclude_KATNTSAP   
76                    Marketing+MBA All India - 240224   
8                All India Generic Leads Ad Set_240224   
74                     MS + Coop KA&TN Ad set - 240224   
90    Open Targeting - Advantage+ audience [ 7th Dec ]   
0                                                   --   
55                                            GMAT/GRE   
11              All India Study abroad Interest 070324   
81                Open Targeting - Advantage+ audience   
2                                                   --   
72                     MS + Coop AP&TS Ad set - 240224   
1                                                   --   
86   Open Targeting - Advantage+ audience [ 31st Oct ]   
83   Open Targeting - Advantage+ audience [ 23rd Nov ]   
13             All India focusing on Uni Ad Set_060324   
19                             Custom-Intent_Targeting   
117                      USP-2_Data_Job-title_KATNTSAP   
77                        Marketing+MBA KA&TN - 240224   

                                         campaign_name  platform  total_spent  \
73                       USP_MS + Coop Campaign_240224  Facebook    647685.15   
5                      USP Study Abroad_Generic_240224  Facebook    535064.20   
116                            USP_2_Lead-gen-1_220124  Facebook    308497.83   
76         USP_Marketing+MBA_New Leads Campaign_240224  Facebook    551614.48   
8                      USP Study Abroad_Generic_240224  Facebook    647092.99   
74                       USP_MS + Coop Campaign_240224  Facebook    478504.92   
90   CBO - USP Masterclass Cold [ 15th Dec ] - High...  Facebook    184929.95   
0                       US_Pathway_2_Lead_gen_1_091123  LinkedIn   1884063.88   
55                          USP_Search_APTNTSKA_250124    Google    689360.86   
11                                  USP_Webinar_070324  Facebook    480984.59   
81       USP_Video_More_Volume_No Logic_India_New Link  Facebook     55359.97   
2                       US_Pathway_2_Lead_gen_3_241123  LinkedIn   1749087.89   
72                       USP_MS + Coop Campaign_240224  Facebook    374850.47   
1                       US_Pathway_2_Lead_gen_2_091123  LinkedIn   1469168.59   
86   CBO - USP Masterclass Cold [ 23rd Nov ] - More...  Facebook    244398.01   
83   CBO - USP Masterclass Cold [ 23rd Nov ] - High...  Facebook     95114.56   
13                     USP Study Abroad_Generic_240224  Facebook    318038.10   
19    USP_Video_APTLTSKA_Lead-Gen_Custom-intent_050324    Google    155920.03   
117                            USP_2_Lead-gen-1_220124  Facebook    176270.83   
77         USP_Marketing+MBA_New Leads Campaign_240224  Facebook    319277.52   

     impressions  clicks  leads  
73       1901299   13861   2341  
5        1516258   10984   1884  
116      1011413    6932   1505  
76       4427087   23048   1486  
8        3341182   16973   1375  
74       1346976    8000   1352  
90        599079   12767   1338  
0        5751255   34923   1232  
55        161249   19360   1100  
11       2839529   11106   1028  
81        308755    5296    950  
2        6317435   36201    874  
72       1258107    5901    753  
1        4225764   26627    723  
86        818847   10820    710  
83        429447    6682    710  
13       1420202    8710    674  
19       1220168   25558    623  
117       656768    3548    619  
77       1959033    7984    562

This details the top 20 ad sets by lead generation, including key metrics such as campaign name, impressions, clicks, and leads. This comprehensive view highlights which ad sets were most effective in generating leads and provides a comparison of their performance across various metrics. By analyzing these details, we can identify high-performing ad sets and understand their contribution to overall campaign success.

In [38]:
# Aggregate data by creative name and calculate the mean CTR
creative_ctr = df.groupby('creative_name')['click_through_rate'].mean().reset_index()

# Sort the data by CTR in descending order and get the top 20
top_creatives_by_ctr = creative_ctr.sort_values(by='click_through_rate', ascending=False).head(20)

# Plot CTR by creative name
plt.figure(figsize=(14, 6))
sns.barplot(x='creative_name', y='click_through_rate', data=top_creatives_by_ctr, palette='viridis')
plt.title('Top 20 Creatives by Click-Through Rate')
plt.xlabel('Creative Name')
plt.ylabel('CTR (%)')
plt.xticks(rotation=90)
plt.show()

The chart displays the top 20 creative names ranked by click-through rate (CTR). It reveals which creatives achieved the highest engagement rates. By showcasing the CTR for these top-performing creatives, the chart allows for a comparison of their effectiveness in driving clicks relative to impressions. This insight helps identify the most successful creative strategies and can inform future design and targeting decisions.

In [39]:
# Aggregate data by creative_name and calculate total spent, impressions, clicks, and leads
creative_metrics = df.groupby(['creative_name', 'campaign_name', 'adset_name','platform'])[['total_spent', 'impressions', 'clicks', 'leads']].sum().reset_index()

# Sort the data by leads in descending order and get the top 20 creatives
top_20_creatives = creative_metrics.sort_values(by='leads', ascending=False).head(20)

# Plot the top 20 creatives by leads
plt.figure(figsize=(14, 8))
sns.barplot(x='creative_name', y='leads', data=top_20_creatives, palette='viridis', order=top_20_creatives['creative_name'].tolist())
plt.title('Top 20 Creatives by Leads')
plt.xlabel('Creative Name')
plt.ylabel('Leads')
plt.xticks(rotation=90)
plt.show()

The chart presents the top 20 creative names ranked by the number of leads generated. It highlights which creatives were most effective in converting interactions into leads. By showing the lead counts for these top-performing creatives, the chart enables comparison of their effectiveness and helps identify which creative names delivered the best results in lead generation. This information can guide future creative strategies and optimize lead acquisition efforts.

In [40]:
# Display details of the top 20 creatives
print("Details of Top 20 Creatives by Leads:")
print(top_20_creatives)

Details of Top 20 Creatives by Leads:
                                          creative_name  \
1147                                  USP_RIT_1_Getpaid   
493       #6 ( Graphic ) - Master's Degree In 12 Months   
1                                                    --   
1333                         USP_kavita&steffi_VideoAd6   
940                             USP_Influencer_videoAd2   
524       #6 ( Graphic ) - Master's Degree In 12 Months   
500       #6 ( Graphic ) - Master's Degree In 12 Months   
504       #6 ( Graphic ) - Master's Degree In 12 Months   
1148                                  USP_RIT_1_Getpaid   
526       #6 ( Graphic ) - Master's Degree In 12 Months   
1146                                  USP_RIT_1_Getpaid   
1014            USP_Program_Linkedin_Static_C11_SB-3YWP   
652                   #8 ( Graphic ) - Study In The USA   
795                                   USP_Drexel_1_EWYL   
1324                         USP_kavita&steffi_VideoAd6   
531       #6 ( Graphic ) - Master's Degree In 12 Months   
410   #4 ( Graphic ) - How To Get Admmission In Your...   
1277                        USP_Webinar_MSunder12Months   
532       #6 ( Graphic ) - Master's Degree In 12 Months   
941                             USP_Influencer_videoAd2   

                                          campaign_name  \
1147                      USP_MS + Coop Campaign_240224   
493   CBO - USP Masterclass Cold [ 15th Dec ] - High...   
1                            USP_Search_APTNTSKA_250124   
1333        USP_Marketing+MBA_New Leads Campaign_240224   
940       USP_Video_More_Volume_No Logic_India_New Link   
524                     USP Study Abroad_Generic_240224   
500   CBO - USP Masterclass Cold [ 23rd Nov ] - High...   
504   CBO - USP Masterclass Cold [ 23rd Nov ] - More...   
1148                      USP_MS + Coop Campaign_240224   
526                     USP Study Abroad_Generic_240224   
1146                      USP_MS + Coop Campaign_240224   
1014                     US_Pathway_2_Lead_gen_1_091123   
652                             USP_2_Lead-gen-1_220124   
795                       USP_MS + Coop Campaign_240224   
1324                    USP Study Abroad_Generic_240224   
531                             USP_2_Lead-gen-1_220124   
410                             USP_2_Lead-gen-1_220124   
1277                                 USP_Webinar_070324   
532                             USP_2_Lead-gen-1_220124   
941       USP_Video_More_Volume_No Logic_India_New Link   

                                             adset_name  platform  \
1147                MS + Coop All India Ad set - 240224  Facebook   
493    Open Targeting - Advantage+ audience [ 7th Dec ]  Facebook   
1                                              GMAT/GRE    Google   
1333                   Marketing+MBA All India - 240224  Facebook   
940                Open Targeting - Advantage+ audience  Facebook   
524            All India Counselling Look-a-like_290224  Facebook   
500   Open Targeting - Advantage+ audience [ 23rd Nov ]  Facebook   
504   Open Targeting - Advantage+ audience [ 31st Oct ]  Facebook   
1148                    MS + Coop KA&TN Ad set - 240224  Facebook   
526               All India Generic Leads Ad Set_240224  Facebook   
1146                    MS + Coop AP&TS Ad set - 240224  Facebook   
1014                                                 --  LinkedIn   
652         USP-2_Data_Job-title_India_exclude_KATNTSAP  Facebook   
795                     MS + Coop KA&TN Ad set - 240224  Facebook   
1324              All India Generic Leads Ad Set_240224  Facebook   
531         USP-2_Data_Job-title_India_exclude_KATNTSAP  Facebook   
410         USP-2_Data_Job-title_India_exclude_KATNTSAP  Facebook   
1277             All India Study abroad Interest 070324  Facebook   
532                       USP-2_Data_Job-title_KATNTSAP  Facebook   
941                              Related To Study In US  Facebook   

      total_spent  impressions  clicks  leads  
1147    363179.88      1208355    8025   1567  
493     175527.01       590020   12490   1275  
1       689360.86       161249   19360   1100  
1333    349155.30      3844267   19526   1016  
940      50556.92       299432    5106    883  
524     264607.03       989837    5357    872  
500      87028.19       414751    6244    658  
504     215547.83       784476   10155    637  
1148    242991.84       858683    3823    607  
526     248570.95      1264680    5037    488  
1146    247045.75       826908    3447    472  
1014    497729.37      1316291    9296    447  
652     101672.76       592552    3269    443  
795     104773.37        89517    1682    424  
1324    198709.93      1662397    8836    418  
531      95464.78       377001    2467    399  
410      46909.52         8535     295    374  
1277    166060.22       923194    3319    368  
532     111897.24       498682    2607    367  
941      43993.45       345301    2598    358

This comprehensive view allows for an in-depth comparison of top-performing creatives, showing their impact across various metrics and helping to identify the most effective strategies for lead generation.

In [41]:
# Plot average impressions by platform
plt.figure(figsize=(10, 6))
sns.barplot(x='platform', y='impressions', data=df, palette='viridis', estimator=np.mean, ci=None)
plt.title('Average impressions by Platform')
plt.xlabel('Platform')
plt.ylabel('Average impressions')
plt.xticks(rotation=45)
plt.show()

C:\Users\KRUSHNA KOUSHIK\AppData\Local\Temp\ipykernel_4080\1048618588.py:3: FutureWarning: 

The `ci` parameter is deprecated. Use `errorbar=None` for the same effect.

  sns.barplot(x='platform', y='impressions', data=df, palette='viridis', estimator=np.mean, ci=None)

The analysis shows that Google ad sets are achieving higher average impressions or reach compared to other platforms. This indicates that Google campaigns are reaching a larger audience on average, suggesting greater visibility and potential engagement.

In [42]:
# Plot leads by platform
plt.figure(figsize=(10, 6))
sns.barplot(x='platform', y='leads', data=df, palette='viridis', estimator=sum, ci=None)
plt.title('Leads by Platform')
plt.xlabel('Platform')
plt.ylabel('Total Leads')
plt.xticks(rotation=45)
plt.show()

C:\Users\KRUSHNA KOUSHIK\AppData\Local\Temp\ipykernel_4080\2669477451.py:3: FutureWarning: 

The `ci` parameter is deprecated. Use `errorbar=None` for the same effect.

  sns.barplot(x='platform', y='leads', data=df, palette='viridis', estimator=sum, ci=None)

Total leads by platform shows that facebook have more because it has more campiagns and all.

In [43]:
# Plot average leads by platform
plt.figure(figsize=(10, 6))
sns.barplot(x='platform', y='leads', data=df, palette='viridis', estimator=np.mean, ci=None)
plt.title('Average Leads by Platform')
plt.xlabel('Platform')
plt.ylabel('Average Leads')
plt.xticks(rotation=45)
plt.show()

C:\Users\KRUSHNA KOUSHIK\AppData\Local\Temp\ipykernel_4080\2387000561.py:3: FutureWarning: 

The `ci` parameter is deprecated. Use `errorbar=None` for the same effect.

  sns.barplot(x='platform', y='leads', data=df, palette='viridis', estimator=np.mean, ci=None)

The analysis indicates that Google ad sets are consistently generating more average leads than those on other platforms. This higher lead generation suggests that Google’s targeting or ad strategies are more effective in converting potential customers into actual leads. The greater average lead count highlights Google's strong performance in attracting and engaging users compared to other advertising platforms. This insight can guide future ad spend decisions and optimization strategies to leverage Google’s effectiveness in lead generation.

In [44]:
# Count the number of campaigns by platform
campaign_counts_by_platform = df.groupby('platform')['campaign_name'].nunique().reset_index()
campaign_counts_by_platform.columns = ['Platform', 'Campaign Count']

# Plot the count of campaigns by platform
plt.figure(figsize=(12, 6))
sns.barplot(x='Platform', y='Campaign Count', data=campaign_counts_by_platform, palette='viridis')
plt.title('Count of Campaigns by Platform')
plt.xlabel('Platform')
plt.ylabel('Number of Campaigns')
plt.xticks(rotation=45)
plt.show()

Facebook has the highest number of campaigns, resulting in a larger number of campaigns with high lead counts. However, due to the large volume of campaigns, the average leads per campaign are lower. This indicates that while Facebook campaigns are numerous and capable of generating many leads overall, the average performance per campaign is reduced compared to platforms like Google, which may have fewer campaigns but higher average leads.

Conclusion¶

Distribution and Skewness: The data shows high activity in December, February, and April. Most campaigns have lower total spends, primarily between $100 and $1000. Similarly, impressions are mostly between 100 and 1000, clicks range from 0 to 100, and leads are often between 0 and 5. Campaigns with '0' leads should be re-evaluated or have their budgets reduced.

Platform Insights: Facebook contributes the majority of data points, significantly influencing overall trends. However, Google ad sets outperform other platforms in generating higher average impressions and leads, showing Google's effectiveness in reaching and converting customers.

Campaign and Creative Analysis: Campaigns with more data points tend to have higher reach and engagement. The top 10 campaigns by lead generation show a strong correlation between total spent and leads, indicating efficient budget use. The top 20 creative names by leads highlight effective strategies with detailed metrics such as campaign name, ad set name, total spent, impressions, clicks, and leads.

Correlation and Relationships: The correlation matrix shows that impressions and clicks have a strong direct relationship, while clicks and leads have no significant relationship. Impressions and total spent, as well as leads and total spent, show slight direct relationships, but clicks and total spent have no significant relationship.

Strategic Insights: The analysis identifies the highest financial investments, showing Google campaigns as more effective on average. Facebook campaigns, despite being numerous, have lower average leads per campaign due to volume. Google's superior lead generation should guide future ad spend decisions and optimization strategies.

Recommendations¶

To optimize campaign performance and maximize ROI, it is essential to focus on improving or reducing the budget for campaigns that have generated '0' leads, ensuring a more efficient use of resources. Given the higher effectiveness of Google campaigns in generating leads, reallocating more budget to these campaigns can enhance overall lead generation. Additionally, it is crucial to enhance strategies for Facebook campaigns to increase their average lead performance, potentially by refining targeting, creative content, and bidding strategies. These insights and conclusions should guide strategic decisions to optimize campaign performance, allocate budgets more effectively, and leverage the unique strengths of each platform for better overall results. By continuously monitoring and adjusting based on performance data, you can ensure sustained improvement and success in your advertising efforts.

In [45]:
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score

# Selecting features and target variable
features = ['clicks', 'total_spent', 'click_through_rate', 'impressions', 'creative_name', 'adset_name', 'campaign_name']
target = 'leads'

# Converting categorical variables into dummy variables
df_dummies = pd.get_dummies(df[features], drop_first=True)

# Splitting the data into train and test sets
X_train, X_test, y_train, y_test = train_test_split(df_dummies, df[target], test_size=0.3, random_state=42)

# Initialize and train the Linear Regression model
lr_model = LinearRegression()
lr_model.fit(X_train, y_train)

# Make predictions
y_pred_lr = lr_model.predict(X_test)

# Evaluate the model
mae_lr = mean_absolute_error(y_test, y_pred_lr)
mse_lr = mean_squared_error(y_test, y_pred_lr)
r2_lr = r2_score(y_test, y_pred_lr)

print(f"Linear Regression:")
print(f"  Mean Absolute Error: {mae_lr}")
print(f"  Mean Squared Error: {mse_lr}")
print(f"  R² Score: {r2_lr}")

Linear Regression:
  Mean Absolute Error: 1.117966062167505
  Mean Squared Error: 7.253360974857454
  R² Score: 0.6144846031324868

Output : The Linear Regression model shows a reasonable predictive performance with an average prediction error of 1.12 leads (MAE) and a manageable average squared error of 7.25 (MSE). The model explains 61.4% of the variance in leads (R²), indicating a moderate to strong fit. Overall, the model effectively captures patterns in lead generation, though there is potential for further improvement.

As Linear Regression did not meet our accuracy expectations, we opted for Random Forest Regression due to its superior ability to handle complex, non-linear relationships. Unlike Linear Regression, which assumes a straight-line relationship between variables, Random Forest Regression leverages an ensemble of decision trees to capture intricate patterns and interactions within the data.

In [48]:
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score




# Selecting features and target variable
features = ['clicks', 'total_spent', 'click_through_rate', 'impressions', 'creative_name', 'adset_name', 'campaign_name']
target = 'leads'

# Converting categorical variables into dummy variables
df_dummies = pd.get_dummies(df[features], drop_first=True)

# Splitting the data into train and test sets
X_train, X_test, y_train, y_test = train_test_split(df_dummies, df[target], test_size=0.3, random_state=42)

# Initialize and train the Random Forest Regressor
rf_model = RandomForestRegressor(random_state=42)
rf_model.fit(X_train, y_train)

# Make predictions
y_pred = rf_model.predict(X_test)

# Evaluate the model
mae = mean_absolute_error(y_test, y_pred)
mse = mean_squared_error(y_test, y_pred)
r2 = r2_score(y_test, y_pred)

print(f"Random Forest Regressor:")
print(f"  Mean Absolute Error: {mae}")
print(f"  Mean Squared Error: {mse}")
print(f"  R² Score: {r2}")

# Feature Importances
importances = rf_model.feature_importances_
feature_names = X_train.columns
feature_importances = pd.Series(importances, index=feature_names).sort_values(ascending=False)

# Get top 20 features
top_20_features = feature_importances.head(20)

print("Top 20 Feature Importances (Random Forest):")
print(top_20_features)

# Plotting the top 20 feature importances
plt.figure(figsize=(12, 8))
sns.barplot(x=top_20_features, y=top_20_features.index)
plt.title('Top 20 Feature Importances (Random Forest)')
plt.xlabel('Importance Score')
plt.ylabel('Feature')
plt.show()

Random Forest Regressor:
  Mean Absolute Error: 0.5375668730288518
  Mean Squared Error: 3.630365880154187
  R² Score: 0.8070464233183396
Top 20 Feature Importances (Random Forest):
clicks                                                                                   0.476358
total_spent                                                                              0.214003
click_through_rate                                                                       0.151406
impressions                                                                              0.033867
creative_name_USP_kavita&steffi_VideoAd6                                                 0.016300
campaign_name_USP Masterclass Cold [ 14th Dec ] - Link Click Ad                          0.012111
adset_name_Open Targeting - Advantage+ audience                                          0.009721
adset_name_Study Abroad                                                                  0.007942
adset_name_MS + Coop All India Ad set - 240224                                           0.006523
creative_name_#4 ( Graphic ) - How To Get Admmission In Your Dream Universities In US    0.005952
campaign_name_USP_MS + Coop Campaign_240224                                              0.004765
campaign_name_USP_Search_India-ex-APTNTSKA_250124                                        0.004203
adset_name_All India Counselling Look-a-like_290224                                      0.003028
adset_name_Open Targeting - Advantage+ audience [ 7th Dec ]                              0.002975
creative_name_USP_Drexel_1_EWYL                                                          0.002519
creative_name_USP_RIT_1_Getpaid                                                          0.002288
campaign_name_USP_2_Lead-gen-1_220124                                                    0.002225
campaign_name_USP Masterclass Cold [ 23rd Nov ] - More volume - With Logic - India       0.002072
creative_name_USP_Program_Linkedin_Static_C11_SB-3YWP                                    0.001965
campaign_name_USP Study Abroad_Generic_240224                                            0.001601
dtype: float64

Conclusion :

The Random Forest Regressor has demonstrated strong predictive performance with an R² score of 0.81, indicating that it effectively captures 81% of the variance in lead generation. The model’s Mean Absolute Error (0.54) and Mean Squared Error (3.63) suggest that predictions are reasonably accurate with manageable error levels. Key insights reveal that the most influential factors for generating leads are the number of clicks and total spending, highlighting their crucial roles in campaign effectiveness. The click-through rate also plays a significant role, albeit less impactful. These findings suggest that the company should focus on optimizing click volumes and strategic spending to maximize lead generation. Additionally, refining ad creatives and campaigns based on their impact can further enhance performance. Regular model updates and feature analysis will help sustain and improve lead prediction accuracy, guiding more informed decision-making and strategic planning.

In [ ]: